“Create a recipe with lentils and chocolate” That’s what my colleagues and I asked Chat GPT during a session to understand what it could and could not do. We thought we were being funny.
But before I get into the details of what it created, let me tell you another tale about Chat GPT. Don't be fooled by its friendly, polite tone. It can have an attitude.
I asked it to explain how antioxidants are increased during the caramelization cooking process.
And like a know-it-all that doesn’t know what it’s talking about, it said: “Caramelization itself absolutely does not increase the antioxidant content of foods”.
I responded by suggesting it should read a research study I provided that explained that this does occur even when caramelizing sugar.
How did it respond? “Thank you for sharing the article. It seems recent research suggests that caramelization may indeed lead to the formation of antioxidants.”
That’s right Chat GPT, you were wrong.
It didn't apologize but in my mind, I think it did. Maybe I’m giving it too much personality.
This is an example of the limitations of AI – it doesn’t know everything. It’s not up-to-date. No one should take what it says as gospel. Check other sources. AI can be very useful in saving time for many things but you have to know what to use it for and what limitations it has.
But let’s get back to the chocolate and lentil recipe. Chap GPT took us seriously. And this is what it created:
Decadent Lentil Chocolate Brownie
1 cup cooked lentils, cooled (green or brown)
1/2 cup unsweetened applesauce
1/4 cup coconut oil
1/2 cup maple syrup
1 teaspoon vanilla extract
1/2 cup whole wheat flour or almond flour
1/2 cup unsweetened cocoa powder
1/2 teaspoon baking powder
1/4 teaspoon salt
1/2 cup dark chocolate chips (optional, for extra richness)
As a food professional and baker who has lived in the world of “healthy” for quite some time, I recognize a recipe that’s trying to be all healthy things to match all health trends, going back 40 years.
Back in the 80s, applesauce was a common substitute for fat – this recipe has halved the amount of fat that would be typical in a brownie and replaced it with applesauce – just as it would have been done in the past.
If you think this is going to lower the taste and texture significantly, you are right. Aren’t you glad the 80s are over? Unfortunately, someone needs to tell Chat GPT.
And it suggests coconut oil, instead of beautiful butter. That’s a more modern health concept and while some people love coconut oil, there’s no reason to avoid butter.
Maple syrup is replacing white sugar, which is fine, but the amount is 75% less than a brownie should have.
There are no eggs so it can be vegan. And the flour is either whole wheat (whole grain) or almond flour (gluten-free)
Did it miss any trend? I don’t think so.
But here’s the question I have. Who asked it to do it this way? We didn’t. How was the data loaded to make it think this was what we were going for? Was it the lentils?
And I love how it named them “decadent”. Only a robot, that doesn’t eat, would think this recipe is decadent.
I was so put off by the lack of tasty ingredients, that I asked it to do it again but with butter and eggs and to get rid of the applesauce. I love homemade applesauce but the stuff from the store? No, thanks.
Chat GPT cannot know that store-bought applesauce doesn’t go with the word “decadent”
This is recipe #2:
1 cup cooked and mashed lentils (green or brown lentils work well)
1/2 cup unsalted butter, melted
1 cup granulated sugar
2 large eggs
1 teaspoon vanilla extract
1/2 cup all-purpose flour
1/3 cup unsweetened cocoa powder
1/4 teaspoon baking powder
1/4 teaspoon salt
1/2 cup chocolate chips (optional, for extra indulgence)
Except for the lentils, the butter, sugar, egg, and flour proportions are standard for a brownie. And whatever health decision it thought it was making in the first version went out the window just because I asked for another version with butter and eggs and no applesauce.
It clearly can’t grasp the nuance of re-creating a recipe with a couple of changes.
I tried to get it to put back the maple syrup and lose the white sugar. But it would take something else out. And it kept changing the amounts that would skew the success of the recipe. I know Chat GT thought it knew what it was doing. But as a professional baker, I knew it didn’t.
There are plenty of brownie recipes with chocolate and lentils – so it’s not new. Is this where Chat GPT got the recipe?
I asked it if the recipes it creates are unique.
It answered: “I generate recipes based on common culinary knowledge and techniques”
At least that is what Chat GPT thinks. I don’t think so. I doubt that they had a team of chefs imputing the data.
This is where a robot fails. It does not eat. It cannot taste, smell or touch. It knows nothing about mouth-feel, crumb structure, or how it feels when flavor sends messages to the brain.
It only knows what it has been told. And cooking and baking require “doing”, not “telling” to get a good recipe.
For example, it doesn’t know that whole wheat flour and almond flour cannot be interchanged 1:1. Only someone who has worked with whole wheat and almond flour would know this.
I did make the first brownie recipe – I was curious about the flavor and texture.
It was a bit gooey, not very tasty and the amount of cocoa gave it a bitter aftertaste. Again, Chat GPT did not know that the word “cocoa” doesn’t indicate good-quality cocoa.
Or that the best brownies are made with chocolate (generally, unsweetened). And if using cocoa, 22/24 is better as it has a higher fat content and therefore, a richer chocolate flavor and is less bitter than the 10/12 type that is commonly found in most stores.
It was also flatter than the brownies I made with eggs (see picture). Chat GPT didn’t know to replace the egg with something like a ground flax-water or chia-water combination. Both can help replace the texture provided by an egg and give it some of the lift that eggs provide.
A couple of days later, I asked Chat GPT again to create a recipe with lentils and chocolate. It had learned. In its mind, I guess it felt it had utterly failed with the brownies since we kept asking it to do it again and again. Yes, chat GPT learns from you and about you.
So, it went in a different direction and gave me a chocolate lentil chili recipe. Again, it did not invent the recipe. There are many chocolate chili recipes on the internet including ones with lentils.
This recipe intrigued me. I’m not a chili fan. But I liked the spices listed and was curious to see how chocolate would fit the recipe.
I don’t know what Chat GPT was thinking as it made it vegan. Is it dedicated to providing more plant-based recipes? Who told it to do that? Or does it think that’s what I want?
I decided to make the recipe but with changes. It had the usual bland, flatulence-inducing kidney beans which I rejected and replaced with roasted cashews. I also added honey because the chocolate and cocoa made it too bitter. As I said, a robot has no sense of taste.
A few other tweaks and I had it more to my taste preference. The recipe is below.
I made it a second time and went for more chocolate and no cocoa. Much better although it’s not something I feel I need to make again.
But I also felt it was missing something - it needed more umami flavor. This could have been accomplished with the addition of meat. fish or perhaps mushrooms to help it keep vegetarian.
I liked it more when I added raw milk, aged cheddar cheese, and full-fat yogurt. I like to drain my yogurt a bit to make it thicker and have a consistency similar to sour cream.
But this is what a robot can’t do. I can make changes to a recipe and know what I’m doing. I can make additions and substitutions because I know what foods taste like and what flavors might work together.
AI is never going to be able to match what we do. And I would hate to live in a world where it’s an accepted practice to have robots create recipes with no human input adjusting the ingredients and proportions.
However, I’m sure there will be people who don’t cook who will use Chat GPT to create recipes with no understanding they may be circulating inferior recipes. But we can spread the word and let the world know that recipes are best when people do the creating and not AI.
Chocolate Lentil Chili
1/2 cup dry brown or green lentils, rinsed and drained
1 medium onion, chopped
3 cloves garlic, minced
1 bell pepper, seeded and cut into thin 2-inch strips
2 carrots, sliced
1 can (14 oz) diced tomatoes
1 can (15 oz) black beans, drained and rinsed
1/2 cup whole cashews, lightly roasted
4 cups vegetable broth
2 tablespoons olive oil
1 tbsp chili powder
1 tsp cumin
1/2 tsp smoked paprika
Salt and pepper to taste
1/2 cup semi-sweet chocolate chips or dark chocolate chunks (70% cocoa or higher)
1 tbsp raw honey
In a pot, combine the rinsed lentils with 1 1/2 cups of vegetable broth, 1 tsp mustard, and 1/2 tsp sea salt. Bring to a boil, then reduce heat to low, cover, and simmer for 20-25 minutes or until lentils are tender but not mushy. Drain any excess liquid. Rinse and let cool.
Heat olive oil in a medium pot over medium heat. Add chopped onion, bell pepper, and carrot. Sauté until the vegetables are softened. Add the garlic chili powder, cumin, smoked paprika and stir for about 30 seconds. Add the tomatoes and broth. Stir and bring to a boil. Lower to simmer and cook for 15 minutes. Add the dark chocolate chips or chunks and the honey. Stir until melted and combined. Add the cooked lentils and black beans. Simmer for 5 minutes. Season with sea salt and pepper to taste.
Tips:
1. Add full-fat yogurt to mellow up to spicy and bitter tastes.
2. Raw milk, aged, cheddar cheese increase the flavor profile. Garnish with green onions to increase the eye appeal.
3. Use organic ingredients as much as possible.
4. Using a good-quality, pure chocolate will work best
It's interesting. One of my colleagues tok a course and they rold her it gets to know you. I ask it alot of questions in the health arena so that may have influenced it. But I have never asked for anything that would indicate a preference for plant-based only. I would love to know how the data is assembled.
Fascinating how Chat GTP decided it should skew things to plant-based?? I suspect if it is going by what is commonly searched for or popular right now maybe that would be the result?